The development of an IP-enabled imaging apparatus, such as a Web camera, makes it possible to construct a wide and large-scale monitoring system using an IP-based network. In addition, a reduction in the cost of a hard disk device and an increase in the capacity of the hard disk device make it possible to store images captured by the monitoring system for a long time. In recent years, there is an increasing demand for a search technique capable of searching desired data from the stored image data.
As a technique for searching images, there is a similarity search technique based on image features. In the search technique, an image feature, which is a numerical vector representing information, such as a color and a shape, is extracted from a frame image of a video, which is a search target, and the image feature is stored as retrieval data in a database. During a search, a distance between the image feature extracted from the image selected by the user and the image feature stored in the database in a vector space is evaluated. An image having a short distance between the image features is determined as a similar image, and is returned as the search result.